Deep Meditations (2018)
Deep Meditations is a series of abstract films and installations that explore how we make meaning and the gap between felt experience and representation. The works invite us on a spiritual journey through slow, meditative, continuously evolving images and sounds: fragments of our collective experience refracted through a deep neural network.
“It feels like the machine is trying to tell me something…” – audience member
What does love look like? What does pain look like? What does faith look like?
You know what love is, not because someone described it to you, but because you felt it first, in your body. Then someone gave that feeling a name: “that which you are feeling right now, that is Love”. And you learnt to associate the word with your inner state.
The experience of love or pain can only be known from within. That felt quality cannot be objectively described and communicated to someone --- or something --- that has no prior experience of it. It cannot be encoded, programmed into, or even taught to, a machine. From the outside, only the traces it leaves in the world can be observed.
The Cloud is full of such traces of experience, archived by the Keepers of our collective consciousness. I trained a generative AI model on these fragments --- hundreds of thousands of images from Flickr, tagged with words like love, pain, faith, grief, joy, and many others. But the model didn’t learn what our inner experiences feel like; it learned what we think they look like.
Training on others’ images is now a major legal and ethical debate, but at the time it wasn’t. It was important to me to initiate these discussions and use only CC0 public-domain images.
Generative AI animations at the time were typically random morphs. Deep Meditations was among the earliest to explore the latent space of generative models as a new medium for storytelling with meaningful narrative control. I developed tools to control how images evolved over time, enabling works with intentional narrative structure. I presented this work at NeurIPS, a leading AI conference.
Sound is generated by another artificial neural network using a custom architecture GranNMA which I trained on hours of religious and spiritual chants, prayers and rituals from around the world.
“We are all connected. To each other, biologically. To the earth, chemically. To the rest of the universe atomically.”
– Neil deGrasse Tyson
“The cosmos is within us. We are made of star-stuff. We are a way for the universe to know itself.”
– Carl Sagan
“How can we think in times of urgencies without the self-indulgent and self-fulfilling myths of apocalypse, when every fiber of our being is interlaced, even complicit, in the webs of processes that must somehow be engaged and repatterned? Recursively, whether we asked for it or not, the pattern is in our hands.
…
It matters what ideas we use to think other ideas with. It matters what matters we use to think other matters with. It matters what knots knot knots.
It matters what thoughts think thoughts. It matters what worlds world worlds”
– Donna Haraway
Research: Latent Storytelling & Narrative
The work is also an exploration into the use of Deep Generative Neural Networks as a medium for creative expression and story telling with meaningful human control over narrative. While the explorations of these so-called ‘latent spaces’ in generative neural networks are typically random, here, precise journeys are carefully constructed in these high dimensional spaces to create a narrative.
Some writing on the technical aspects:
- I presented some of the techniques at the 2nd Workshop on Machine Learning for Creativity and Design at the Neural Information Processing Systems (NeurIPS) 2018 conference, one of the most prestigious academic conferences on artificial intelligence. Paper: https://nips2018creativity.github.io/doc/Deep_Meditations.pdf
- I expand on the paper here
- I discuss the work in depth in my PhD Thesis.
- Code: https://github.com/memo/py-msa-kdenlive (note this code does not produce these artwork’s. It’s just a tool I developed and used to create the artworks)
Credits & Acknowledgments
Created during my PhD, funded by the EPSRC.
Visual network is a ProGAN.
Audio network is a custom architecture GranNMA.
Amongst many people, I’d like to especially thank Nina Miolane and Sylvain Calinon for their contributions and help with Riemannian Geometry, and Miolane et al for geomstats.
Series
InstallationVideoArtificial IntelligenceEcologyGalleryMusic & Sound DesignConsciousnessMindOpen Source
VideoArtificial IntelligenceEcologyConsciousnessGalleryMindMusic & Sound DesignOpen Source
PerformanceVideoArtificial IntelligenceEcologyGalleryConsciousnessMindMusic & Sound DesignOpen Source
VideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
VideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
Selected Exhibitions, Performances & Presentations(10 / 10)
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryLanguageMindMusic & Sound DesignSimulationOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
PerformanceVideoArtificial IntelligenceEcologyGalleryConsciousnessMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source
InstallationVideoArtificial IntelligenceConsciousnessEcologyGalleryMindMusic & Sound DesignOpen Source